Knowledge graph construction method and system of hydrothermal mineralization system

Abstract

The invention discloses a method and system for constructing a knowledge graph of a hydrothermal metallogenic system, and the method comprises the steps: constructing a BERT-BiLSTM-CRF semantic extraction model, and inputting the geological text data of a target region into the semantic extraction model; extracting context features of the input text based on a BERT algorithm, and obtaining a word sequence of each sentence of the input text; forward information and backward information in the word sequence are obtained through a bidirectional long-short term memory network, and key word information is extracted; marking the key word information through a conditional random field, and outputting geological key data in a triple storage format; according to the triple storage format of the geological key data, map nodes are set, and a knowledge map framework is constructed; and arranging the geological key data in a knowledge graph framework to form a geological knowledge graph. A knowledge graph is constructed based on bidirectional processing data and a triple format, planned and structured data expression is realized, and refined expression output of hydrothermal metallogenic geologic features is realized.

Description

Technical Field

[0001] The present invention relates to the field of geological mineral technology, and in particular to a knowledge graph construction method and system for a hydrothermal mineralization system. Background Art

[0002] Knowledge graphs can organize massive amounts of knowledge data into structured knowledge expressions, which can intuitively reflect the data content and help obtain the hidden relationships between data. Due to the complexity of hydrothermal mineralization, the multi-source heterogeneity of data, and the data chaos, the existing prediction and analysis accuracy of hydrothermal mineralization systems is low. How to organize and express the data of hydrothermal mineralization systems is one of the important research directions for predictive analysis of hydrothermal mineralization systems. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art. The present invention provides a knowledge graph construction method and system for a hydrothermal mineralization system. The key features of the data are obtained by bidirectional processing of text data, stored in a triple format and the knowledge graph nodes are set according to the triple specifications, so as to realize planned and structured data expression and realize the refined expression and output of the hydrothermal mineralization geological characteristics.

[0004] The present invention provides a method for constructing a knowledge graph of a hydrothermal mineralization system, the method comprising:

[0005] Construct a BERT-BiLSTM-CRF semantic extraction model, and input the geological text data of the target area into the semantic extraction model;

[0006] Extract the context features of the input text based on the BERT algorithm and obtain the word sequence of each sentence of the input text;

[0007] Obtaining forward information and backward information in the word sequence through a bidirectional long short-term memory network BiLSTM to extract key word information;

[0008] The key word information is marked by the conditional random field (CRF), and the key geological data is output in a triple storage format;

[0009] Set graph nodes according to the triple storage format of key geological data and build a knowledge graph framework;

[0010] Organize key geological data within the knowledge graph framework to form a geological knowledge graph.

[0011] Furthermore, the constructing of the BERT-BiLSTM-CRF semantic extraction model and inputting the geological text data of the target area into the semantic extraction model comprises:

[0012] The BERT-BiLSTM-CRF semantic extraction model is constructed by combining the semantic training model BERT, the bidirectional long short-term memory network BiLSTM and the conditional random field CRF.

[0013] Query the text data corresponding to the target area in the database to obtain the geological text data of the target area;

[0014] The geological text data of the target area is input into the BERT-BiLSTM-CRF semantic extraction model to obtain the key geological data of the target area.

[0015] Furthermore, the extracting of context features of the input text based on the BERT algorithm to obtain a word sequence of each sentence of the input text includes:

[0016] Based on the self-attention mechanism, the semantic influence of all words in the input text on the current word is obtained, and the contextual features of the input text are extracted;

[0017] Based on the feedforward neural network, the semantic features are nonlinearly transformed and the word sequence is output.

[0018] Furthermore, the step of obtaining forward information and backward information in the word sequence through a bidirectional long short-term memory network BiLSTM and extracting key word information includes:

[0019] The word order is processed forward to obtain the forward vector result, and the word order is processed reversely to obtain the reverse vector result;

[0020] Combine the forward vector results with the reverse vector results to obtain the key word information in the word order.

[0021] Furthermore, the step of obtaining forward information and backward information in the word sequence through a bidirectional long short-term memory network BiLSTM and extracting keyword information further includes:

[0022] Match the extracted key word information with the relevant word descriptions recorded in the database to extract the corresponding description sentences;

[0023] Semantic analysis is performed based on the description sentence, and semantic verification is performed on the key word information.

[0024] Furthermore, the construction method also includes:

[0025] By batch reading the data files with CRF tags as the dataset, and dividing the dataset into training set and test set in proportion;

[0026] The training set is input into the BERT-BiLSTM-CRF semantic extraction model for training, and the trained semantic extraction model is tested and verified based on the test set.

[0027] Furthermore, the geological key data is organized in the knowledge graph framework to form a geological knowledge graph, including:

[0028] A geological knowledge map of the hydrothermal mineralization system in the target area is constructed, and the graph database algorithm is used to mine the correlation between the data of each node of the geological knowledge map.

[0029] Furthermore, the geological knowledge map of the hydrothermal mineralization system in the target area is constructed, and the association of each node data of the geological knowledge map is mined by applying a graph database algorithm, including:

[0030] Perform centrality analysis on each node data in the knowledge graph:

[0031] PR(A)=(1-d)+d*∑PR(T) / C(T);

[0032] Among them, PR(A) represents the Page Rank value of the current node data A, PR(T) represents the Page Rank value of other node data T linked to the current node data A, C(T) represents the number of outbound links of node data T, and d is the damping factor.

[0033] Furthermore, the geological knowledge map of the hydrothermal mineralization system in the target area is constructed, and the association of each node data of the geological knowledge map is mined by applying a graph database algorithm, including:

[0034] Perform community detection on each node data of the geological knowledge graph:

[0035] ΔQ=(Sum(in(Ci)) / m-Sum(tot(Ci)) 2 / 2m 2 )-(Sum(in(C)) / m-Sum(tot(C) 2 / 2m 2 );

[0036] Among them, ΔQ represents the degree of improvement of the community assignment of the node, in(Ci) represents the sum of the incoming edge weights of all nodes in community Ci, tot(Ci) represents the sum of the degrees of all nodes in community Ci, C represents the current community where the node is located, m represents the number of edges in the graph, and sum is the summation symbol.

[0037] The present invention also provides a knowledge graph system for a hydrothermal mineralization system, wherein the knowledge graph system is formed based on the construction method, and the knowledge graph system comprises:

[0038] BERT-BiLSTM-CRF semantic extraction model: used to extract geological text data of the target area and output the geological key data of the target area in triple storage format;

[0039] Framework construction module: used to set graph nodes according to the triple storage format of geological key data and build a knowledge graph framework;

[0040] Knowledge graph generation module: organizes key geological data within the knowledge graph framework to form a geological knowledge graph.

[0041] The embodiment of the present invention provides a method and system for constructing a knowledge graph of a hydrothermal mineralization system. By setting a data storage mode of triple specifications and cooperating with a knowledge graph framework for planned and structured data storage, in the knowledge graph, by calculating the graph database algorithm on the node data, it is possible to achieve accurate analysis and organization of the mineralization data in the knowledge graph, thereby improving the accuracy of the prediction of the mineralization laws of the target area. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a flow chart of a method for constructing a knowledge graph of a hydrothermal mineralization system in an embodiment of the present invention;

[0044] Figure 2 It is a framework diagram of text data processing in an embodiment of the present invention;

[0045] Figure 3 1 is a diagram of a text data feature extraction and storage framework in an embodiment of the present invention;

[0046] Figure 4 It is a framework diagram of knowledge graph construction in an embodiment of the present invention;

[0047] Figure 5 It is a schematic diagram of the knowledge graph system of the hydrothermal mineralization system in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Embodiment 1:

[0050] Figure 1A flow chart of a method for constructing a knowledge graph of a hydrothermal mineralization system in an embodiment of the present invention is shown, and the construction method includes:

[0051] S11: Build a BERT-BiLSTM-CRF semantic extraction model, and input the geological text data of the target area into the semantic extraction model.

[0052] The BERT-BiLSTM-CRF semantic extraction model is constructed by combining the semantic training model BERT, the bidirectional long short-term memory network BiLSTM and the conditional random field CRF.

[0053] Query the text data corresponding to the target area in the database to obtain the geological text data of the target area;

[0054] The geological text data of the target area is input into the BERT-BiLSTM-CRF semantic extraction model to obtain the key geological data of the target area.

[0055] Furthermore, the semantic training model (BERT, Bidirectional Encoder Representations from Transformers) is a bidirectional encoder representation algorithm that can analyze the semantic order of text content to extract characteristic language data.

[0056] Furthermore, the bidirectional long short-term memory network (BiLSTM) performs bidirectional analysis operations on text semantics and obtains feature information in text semantics by forward and reverse processing of text data.

[0057] Furthermore, Conditional Random Field (CRF) is a statistical modeling method used to model random variables in structured data. In the field of machine learning, it is often used for sequence labeling tasks in natural language processing, such as part-of-speech tagging and named entity recognition.

[0058] S12: Extract the context features of the input text based on the BERT algorithm and obtain the word sequence of each sentence of the input text.

[0059] Specifically, the BERT algorithm obtains the semantic influence of all words in the input text on the current word based on the self-attention mechanism, extracts the contextual features of the input text, and performs nonlinear transformation on the semantic features based on a feedforward neural network to output a word sequence.

[0060] The BERT algorithm is used to perform semantic analysis on the input text. Based on bidirectional feature analysis, the semantic model can perform full-text analysis based on the sentence semantics of the input text, thereby improving the accuracy of the semantic analysis of the input text. It can extract enough text words from massive text content and arrange them in the semantic order of the input text, thereby outputting a word sequence.

[0061] S13: Obtain forward information and backward information in the word sequence through a bidirectional long short-term memory network BiLSTM, and extract key word information.

[0062] Specifically, the word order is processed forward to obtain a forward vector result, and the word order is processed reversely to obtain a reverse vector result. The forward vector result and the reverse vector result are combined to obtain the key word information in the word order. Based on the comprehensive analysis of the forward vector result and the reverse vector result, the feature data in the text data can be extracted and matched with the corresponding semantic feature type.

[0063] Furthermore, in the BiLSTM algorithm, by inputting the text (h0, h1, h2) in the forward LSTML, the vector {hL0, hL1, hL2} is obtained; in the backward LSTMR of the BiLSTM algorithm, the text (h0, h1, h2) is input, and three vectors {hR0, hR1, hR2} are obtained. Finally, the forward and backward latent vectors are concatenated to obtain {[hL0, hR2], [hL1, hR1], [hL2, hR0]}, that is, {h0, h1, h2}. For the sentiment classification task, the sentence representation of the input text is often [hL2, hR2], which contains all the information in the forward and backward directions, so that the feature data in the sentence can be extracted.

[0064] S14: Label the key word information through the conditional random field (CRF) and output the geological key data in a triple storage format.

[0065] By marking key word information based on conditional random fields and semantically sorting and extracting key words from text data, the accuracy of word extraction from text data can be improved.

[0066] Furthermore, Conditional Random Field (CRF) is a statistical modeling method used to model random variables in structured data. In the field of machine learning, it is often used in sequence labeling tasks in natural language processing, such as part-of-speech tagging, named entity recognition, etc., and can effectively discover the arrangement rules of text data features of key word information.

[0067] S15: Set graph nodes according to the triple storage format of geological key data and build a knowledge graph framework.

[0068] Based on Neo4j, the data in triple format is structured and normalized, that is, a knowledge graph of the target area is constructed, and the nodes of the knowledge graph are constructed based on the setting rules of the triple format. In this embodiment, the mineral deposit name, relationship component, and relationship attribute are used as nodes of the knowledge graph to construct a knowledge graph framework, and the triple data is imported into the knowledge graph framework to achieve planned storage of the triple data.

[0069] S16: Organize key geological data within the knowledge graph framework to form a geological knowledge graph.

[0070] Specifically, the centrality, similarity, community detection and link prediction analysis of the node data of the knowledge graph framework are performed through the graph database algorithm to obtain the correlation between the data of each node in the knowledge graph, that is, the geological characteristic data of the target area, thereby improving the accuracy of the knowledge graph's mineralization prediction analysis of the target area.

[0071] An embodiment of the present invention provides a method for constructing a knowledge graph of a hydrothermal mineralization system. By setting a data storage mode of triple specifications and cooperating with a knowledge graph framework for planned and structured data storage, in the knowledge graph, by calculating the graph database algorithm on the node data, it is possible to achieve accurate analysis and organization of the mineralization data in the knowledge graph, thereby improving the accuracy of the prediction of the mineralization laws of the target area.

[0072] Embodiment 2:

[0073] Figure 2 The framework diagram of text data processing in the embodiment of the present invention is shown. According to the location of the target area to be analyzed, the text data of the corresponding target area is extracted from the database.

[0074] Specifically, the database includes relevant project books, paper documents, geological map data, etc. of the target area. Based on keyword retrieval, text data related to the target area can be quickly obtained in the database, and simple data preprocessing and classification are performed on the obtained text data, that is, the text data is normalized.

[0075] Furthermore, the acquired text data is subjected to sentence extraction and organization, and geologically relevant keyword retrieval is performed based on the target area. In the present embodiment, the text data of the target area is normalized according to categories such as gravity and magnetic anomaly data, chemical anomaly data, and mineralization system characteristic data of the target area. The organized text data is divided into data sets and used as analysis data sets of the semantic analysis model to extract corresponding geological characteristic data based on the organized text data sets.

[0076] Furthermore, with respect to the processing of the image data, based on the analysis of the geological feature contours in the image data and the image grayscale values, the geological features of the target area are matched accordingly, thereby outputting and organizing the geological feature data of the target area.

[0077] Furthermore, according to the geographical characteristics of the target area, the target knowledge types are obtained, such as the geological map of the target area, including the distribution of strata and lithology, etc.; the mineralization system information of the target area: including the National 305 Project series and the literature of the study area, and the East Tianshan mineralization belt map drawn by previous studies; the gravity and magnetic anomaly and chemical anomaly data of the target area: the gravity and magnetic anomaly and chemical anomaly data related to the ore deposit.

[0078] Specifically, Figure 3 The text data feature extraction and storage framework diagram in the embodiment of the present invention is shown, the acquired text data of the target area is analyzed, extracted and stored, the geological feature data of the target area is extracted from the text data, and the geological feature data is stored in a triple format. In this embodiment, the mineral deposit information of the target area is sorted and stored, and the format of "ore deposit name, relationship component, relationship attribute" is used for storage, which can intuitively reflect the characteristic description of the mineralization system of the target area.

[0079] Furthermore, the name of the mineral deposit is: all the magmatic hydrothermal mineral deposits in the target area are taken as the head entity, and the corresponding mineral deposit names in the target area are extracted according to the mineral deposit data recorded in the text data.

[0080] Relationship components: mainly summarize 19 categories around the characteristic description of mineral deposits, including mineral type, ore-bearing stratum, material source, fluid source, heat source, alteration type, mineralization era, tectonic unit, stratum name, surrounding rock, rock genetic type, rock series, sedimentary structure type, lithology description, fossil combination, geotectonic environment, rock structure, rock structure, isotope age, deposit type, etc.

[0081] Relationship attribute: refers to the corresponding relationship description of geological minerals of different deposit types.

[0082] Specifically, the BERT algorithm is used to extract features from text data. Based on bidirectional feature analysis, the semantic model can perform full-text analysis according to the sentence semantics of the input text, thereby improving the accuracy of the semantic analysis of the input text. It can extract enough text words from massive text content and arrange them in the semantic order of the input text, thereby outputting a word sequence.

[0083] The BiLSTM network model is used to perform bidirectional vector analysis on the word sequence. By performing forward and reverse processing on the word sequence, a combined analysis is performed based on the word feature vectors of the forward processing and the word feature vectors of the reverse processing to obtain the key word information in the word sequence.

[0084] Specifically, the extracted key word information is matched with the relevant word descriptions recorded in the database to extract the corresponding description sentences. Each key word information is sequentially input into the database for retrieval, and the description sentences corresponding to the key words are extracted from the recorded text content of the database. Semantic analysis is performed based on the description sentences, and semantic verification is performed on the key word information, that is, the semantic content of the key words is obtained, thereby improving the accuracy of text data sorting and classification.

[0085] Furthermore, the description sentences are retrieved in a predetermined sentence format, and simple sentence patterns are set to retrieve sentences related to the keywords, which can accurately extract the definitional sentence patterns of the keywords, so as to detect and analyze the semantics of the keywords according to the definitional sentence patterns, thereby achieving accurate classification of the keywords.

[0086] Furthermore, the entity information data is labeled and classified based on the conditional random field, and the data is stored in a triple format.

[0087] The entity information of the text data is annotated based on the BIO annotation strategy. The letter "B" is used to annotate the beginning of each entity information, the letter "I" is used to annotate the middle to the end of the entity information, and the letter O is used to annotate other text information content positions.

[0088] Table 1: Example of labels for the mineral deposit corpus based on BIO annotation

[0089]

[0090] Among them, MIN stands for mineral deposit and ERA stands for mineralization era. The accuracy of BIO identification can be improved based on sufficient text data.

[0091] Specifically, build a BERT-BiLSTM-CRF model, train the BERT-BiLSTM-CRF model based on the BIO annotated text data, batch acquire the processed text data through the batch acquisition script, build a data set based on the acquired text data, and divide the data set into a training set and a test set, train and debug the BERT-BiLSTM-CRF model based on the training set, and detect whether the trained BERT-BiLSTM-CRF model meets the actual usage requirements based on the test set.

[0092] Specifically, Figure 4 A framework diagram for constructing a knowledge graph in an embodiment of the present invention is shown, in which a knowledge graph of the mineralization system of the target area is constructed, and graph database algorithms (such as centrality, similarity, community discovery, etc.) are applied to mine implicit associations in the mineralization system, thereby improving the depth and breadth of research on mineralization laws.

[0093] Specifically, the data in triple format is structured and normalized based on Neo4j, that is, a knowledge graph of the target area is constructed, and the nodes of the knowledge graph are constructed based on the setting rules of the triple format. In this embodiment, the mineral deposit name, relationship component, and relationship attribute are used as nodes of the knowledge graph to construct a knowledge graph framework, and the triple data is imported into the knowledge graph framework to achieve planned storage of the triple data.

[0094] Furthermore, the Neo4j graph database is one of the data structure methods for storing data. A graph database is a database that stores data in the form of a graph structure, and stores application data in the form of nodes, relationships, and attributes.

[0095] Furthermore, the triple data is extracted based on Cypher statements to create a knowledge graph. Cypher statements are a declarative query language for the graph database Neo4j, which extracts information or modifies data by matching nodes and relationships in the graph database. Based on the Cypher statements, nodes, relationships, labels, and attributes can be created, updated, and deleted. In this embodiment, nodes are created based on Cypher statements, and the names of mineral deposits, relationship components, and relationship attribute data are extracted based on the triple data, thereby constructing a knowledge graph of the target area.

[0096] Specifically, the correlation of the target area mineralization data in the knowledge graph is analyzed based on the algorithm in the graph database. In this embodiment, the target area data of the knowledge graph is analyzed by the centrality algorithm, similarity algorithm, community detection algorithm and link prediction algorithm of the graph database.

[0097] Furthermore, the centrality analysis is performed on each node data in the knowledge graph:

[0098] PR(A)=(1-d)+d*∑PR(T) / C(T);

[0099] Among them, PR(A) represents the Page Rank value of the current node data A, PR(T) represents the Page Rank value of other node data T linked to the current node data A, C(T) represents the number of outbound links of node data T, and d is the damping factor.

[0100] The specific implementation of the PageRank algorithm is to simulate the link relationship between each node data through a mathematical model. The PageRank value of the current data is calculated based on the PageRank algorithm. The centrality of the current data and other data, that is, the relative importance, can be obtained to extract the corresponding relevant data information in the knowledge graph.

[0101] Furthermore, the centrality calculation can also be analyzed and calculated using the centrality algorithm (BC, Betweenness Centrality). The nodes with high Betweenness Centrality values ​​play a greater intermediary role in the graph, and the more important these nodes are in the graph. The formula of the Betweenness Centrality algorithm is as follows:

[0102] BC(v)=∑s≠t≠v(δ(s,t|v)) / (n-1)(n-2));

[0103] Among them, BC(v) represents the Betweenness Centrality value of node v, δ(s,t|v) represents the number of nodes v passed by the shortest path from node s to node t, and n represents the total number of nodes in the graph. The formula of the Betweenness Centrality algorithm shows that the Betweenness Centrality value of node v is equal to the average number of the shortest paths between all other node pairs passing through node v. By calculating the distribution spacing between node data, the importance of node data in the knowledge graph is obtained, thereby obtaining the centrality of the node.

[0104] Specifically, similarity analysis is performed on the node data in the knowledge graph, and the calculation formula for the similarity analysis is:

[0105]

[0106] Among them, E and F are two sets, |E∩F| is the intersection of E and F, and |E∪F| is the union of E and F.

[0107] Specifically, community detection is an algorithm used to analyze the relationship between nodes in a graph. In a knowledge graph, the relationship between nodes usually forms several communities. The purpose of the community detection algorithm is to find the community of the knowledge graph and obtain the subordinate relationship between each node and the corresponding community.

[0108] Perform community detection on each node data of the geological knowledge graph:

[0109] ΔQ=(Sum(in(Ci)) / m-Sum(tot(Ci)) 2 / 2m 2 )-(Sum(in(C)) / m-Sum(tot(C) 2 / 2m 2 );

[0110] Among them, ΔQ represents the degree of improvement of the community assignment of the node, in(Ci) represents the sum of the incoming edge weights of all nodes in community Ci, tot(Ci) represents the sum of the degrees of all nodes in community Ci, C represents the current community where the node is located, m represents the number of edges in the graph, and sum is the summation symbol.

[0111] The embodiment of the present invention provides a method for constructing a hydrothermal mineralization knowledge graph, which uses a triple format to organize and store the information of the magmatic-hydrothermal mineralization system in the target area, and realizes structured and standardized data management. The mineral deposit information is stored in a graph database, and the graph database analysis algorithm (such as centrality, similarity, community discovery, etc.) is used to obtain the implicit association of each node data of the mineralization system, thereby improving the depth and breadth of the research on mineralization laws.

[0112] Embodiment three:

[0113] Figure 5 A schematic diagram of a knowledge graph system for a hydrothermal mineralization system in an embodiment of the present invention is shown, and the knowledge graph system for a hydrothermal mineralization system includes:

[0114] BERT-BiLSTM-CRF semantic extraction model 10: used to extract geological text data of the target area and output geological key data of the target area in triple storage format.

[0115] A BERT-BiLSTM-CRF semantic extraction model is constructed, and the geological text data of the target area is input into the semantic extraction model.

[0116] The BERT-BiLSTM-CRF semantic extraction model is constructed by combining the semantic training model BERT, the bidirectional long short-term memory network BiLSTM and the conditional random field CRF.

[0117] Query the text data corresponding to the target area in the database to obtain the geological text data of the target area;

[0118] The geological text data of the target area is input into the BERT-BiLSTM-CRF semantic extraction model to obtain the key geological data of the target area. The context features of the input text are extracted based on the BERT algorithm to obtain the word sequence of each sentence of the input text.

[0119] Specifically, the BERT algorithm obtains the semantic influence of all words in the input text on the current word based on the self-attention mechanism, extracts the contextual features of the input text, and performs nonlinear transformation on the semantic features based on a feedforward neural network to output a word sequence.

[0120] The BERT algorithm is used to perform semantic analysis on the input text. Based on bidirectional feature analysis, the semantic model can perform full-text analysis based on the sentence semantics of the input text, thereby improving the accuracy of the semantic analysis of the input text. It can extract enough text words from massive text content and arrange them in the semantic order of the input text, thereby outputting a word sequence.

[0121] The forward information and backward information in the word sequence are obtained through a bidirectional long short-term memory network BiLSTM to extract key word information.

[0122] Specifically, the word order is processed forward to obtain a forward vector result, and the word order is processed reversely to obtain a reverse vector result. The forward vector result and the reverse vector result are combined to obtain the key word information in the word order. Based on the comprehensive analysis of the forward vector result and the reverse vector result, the feature data in the text data can be extracted and matched with the corresponding semantic feature type.

[0123] Furthermore, in the BiLSTM algorithm, by inputting the text (h0, h1, h2) in the forward LSTML, the vector {hL0, hL1, hL2} is obtained; in the backward LSTMR of the BiLSTM algorithm, the text (h0, h1, h2) is input, and three vectors {hR0, hR1, hR2} are obtained. Finally, the forward and backward latent vectors are concatenated to obtain {[hL0, hR2], [hL1, hR1], [hL2, hR0]}, that is, {h0, h1, h2}. For the sentiment classification task, the sentence representation of the input text is often [hL2, hR2], which contains all the information in the forward and backward directions, so that the feature data in the sentence can be extracted.

[0124] The key word information is marked by the conditional random field (CRF), and the geological key data is output in a triple storage format.

[0125] By marking key word information based on conditional random fields and semantically sorting and extracting key words from text data, the accuracy of word extraction from text data can be improved.

[0126] Framework construction module 20: used to set graph nodes according to the triple storage format of geological key data and construct a knowledge graph framework.

[0127] Knowledge graph generation module 30: organizes key geological data within the knowledge graph framework to form a geological knowledge graph.

[0128] Based on Neo4j, the data in triple format is structured and normalized, that is, a knowledge graph of the target area is constructed, and the nodes of the knowledge graph are constructed based on the setting rules of the triple format. In this embodiment, the mineral deposit name, relationship component, and relationship attribute are used as nodes of the knowledge graph to construct a knowledge graph framework, and the triple data is imported into the knowledge graph framework to achieve planned storage of the triple data.

[0129] Organize key geological data within the knowledge graph framework to form a geological knowledge graph.

[0130] Knowledge graph verification module 40: performs verification on the node data of the knowledge graph in terms of centrality, similarity, community discovery, etc., and outputs the correlation between the node data in the knowledge graph.

[0131] Specifically, the centrality, similarity, community detection and link prediction analysis of the node data of the knowledge graph framework are performed through the graph database algorithm to obtain the correlation between the data of each node in the knowledge graph, that is, the geological characteristic data of the target area, thereby improving the accuracy of the knowledge graph's mineralization prediction analysis of the target area.

[0132] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0133] In addition, the above provides a detailed introduction to the embodiments of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for constructing a knowledge graph of a hydrothermal mineralization system, characterized in that: The construction method comprises: Construct a BERT-BiLSTM-CRF semantic extraction model, and input the geological text data of the target area into the semantic extraction model; Extract the context features of the input text based on the BERT algorithm and obtain the word sequence of each sentence of the input text; Obtaining forward information and backward information in the word sequence through a bidirectional long short-term memory network BiLSTM to extract key word information; The key word information is marked by the conditional random field (CRF), and the geological key data is output in a triple storage format; Set graph nodes according to the triple storage format of key geological data and build a knowledge graph framework; Organize key geological data within the knowledge graph framework to form a geological knowledge graph.

2. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 1, characterized in that: The constructing of the BERT-BiLSTM-CRF semantic extraction model and inputting the geological text data of the target area into the semantic extraction model comprises: The BERT-BiLSTM-CRF semantic extraction model is constructed by combining the semantic training model BERT, the bidirectional long short-term memory network BiLSTM and the conditional random field CRF. Query the text data corresponding to the target area in the database to obtain the geological text data of the target area; The geological text data of the target area is input into the BERT-BiLSTM-CRF semantic extraction model to obtain the key geological data of the target area.

3. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 1, characterized in that: The process of extracting context features of the input text based on the BERT algorithm and obtaining the word sequence of each sentence of the input text includes: Based on the self-attention mechanism, the semantic influence of all words in the input text on the current word is obtained, and the contextual features of the input text are extracted; Based on the feedforward neural network, the semantic features are nonlinearly transformed and the word sequence is output.

4. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 1, characterized in that: The method of obtaining forward information and backward information in the word sequence by using a bidirectional long short-term memory network BiLSTM and extracting key word information includes: The word order is processed forward to obtain the forward vector result, and the word order is processed reversely to obtain the reverse vector result; Combine the forward vector results with the reverse vector results to obtain the key word information in the word order.

5. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 4, characterized in that: The step of obtaining forward information and backward information in the word sequence by using a bidirectional long short-term memory network BiLSTM and extracting key word information further comprises: Match the extracted key word information with the relevant word descriptions recorded in the database to extract the corresponding description sentences; Semantic analysis is performed based on the description sentence, and semantic verification is performed on the key word information.

6. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 1, characterized in that: The construction method also includes: By batch reading the data files with CRF tags as the dataset, and dividing the dataset into training set and test set in proportion; The training set is input into the BERT-BiLSTM-CRF semantic extraction model for training, and the trained semantic extraction model is tested and verified based on the test set.

7. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 1, characterized in that: The said arranging geological key data in the knowledge graph framework to form a geological knowledge graph includes: A geological knowledge map of the hydrothermal mineralization system in the target area is constructed, and the graph database algorithm is used to mine the correlation between the data of each node of the geological knowledge map.

8. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 7, characterized in that: The step of constructing a geological knowledge map of the hydrothermal mineralization system in the target area and mining the relevance of each node data of the geological knowledge map using a graph database algorithm includes: Perform centrality analysis on each node data in the knowledge graph: PR(A)=(1-d)+d*∑PR(T) / C(T); Among them, PR(A) represents the Page Rank value of the current node data A, PR(T) represents the Page Rank value of other node data T linked to the current node data A, C(T) represents the number of outbound links of node data T, and d is the damping factor.

9. The method for constructing a knowledge graph of a hydrothermal mineralization system according to claim 7, characterized in that: The step of constructing a geological knowledge map of the hydrothermal mineralization system in the target area and mining the relevance of each node data of the geological knowledge map using a graph database algorithm includes: Perform community detection on each node data of the geological knowledge graph: ΔQ=(Sum(in(In)) / m-Sum(tot(in)) 2 / 2m 2 )-(Sum(in(C)) / m-Sum(tot(C) 2 / 2m 2 ): Among them, ΔQ represents the degree of improvement of the community assignment of the node, in(Ci) represents the sum of the edge weights of all nodes in the community Ci, tot(Ci) represents the sum of the degrees of all nodes in the community Ci, C represents the current community where the node is located, and m represents the number of edges in the graph. su m is the summation symbol.

10. A knowledge graph system for hydrothermal mineralization system, characterized in that: The knowledge graph system is formed based on the construction method according to any one of claims 1 to 9, and the knowledge graph system includes: BERT-BiLSTM-CRF semantic extraction model: used to extract geological text data of the target area and output geological key data of the target area in triple storage format; Framework construction module: used to set graph nodes according to the triple storage format of geological key data and build a knowledge graph framework; Knowledge graph generation module: organizes key geological data within the knowledge graph framework to form a geological knowledge graph.

Images